Divided Government and the Fragmentation of American Law

Transcription

1 Divided Government and the Fragmentation of American Law Sean Farhang Miranda Yaver We investigate institutional explanations for Congress s choice to fragment statutory frameworks for policy implementation. We argue that divided party government, which fuels legislative-executive conflict over control of the bureaucracy, motivates Congress to fragment implementation power as a strategy to enhance its control over implementation. We develop a novel measure of fragmentation in policy implementation, collect data on it over the period 1947 to 2008, and test hypotheses linking separation of powers structures to legislative design of fragmented implementation power. We find that divided party government is powerfully associated with fragmentation in policy implementation, and that this association contributed to the long-run growth of fragmentation in the post-war U.S. We further find that legislative coalitions are more likely to fragment implementation power in the face of greater uncertainty about remaining in the majority. Associate Professor of Political Science and Public Policy at the University of California, Berkeley. Charles and Louise Travers Department of Political Science. 210 Barrows Hall #1950 Berkeley, CA (farhang@berkeley.edu). Ph.D. Candidate in Political Science at Columbia University. Department of Political Science. 420 West 118th Street, 7th Floor, New York, NY (mey2111@columbia.edu). We thank Tom Clark, Marc Meredith, and Scott Ainsworth for their helpful comments and suggestions. We gratefully acknowledge the support of the National Science Foundation, Law and Social Science Program, Grant No Data and replication code are available through the American Journal of Political Science dataverse: 1

2 In his classic work on American bureaucracy, James Q. Wilson (1989: ) characterized American policy implementation as a barroom brawl with many participants and no referee. He contrasted this fragmented and decentralized American style of policy implementation with a European model in which, according to Wilson, policy implementation is carried out by only a few key authoritative actors in a more centralized and hierarchical system (see also Moe 1989, 1990; Kagan 1999, 2001). This notion that American policy implementation is fragmented has become something of a commonplace among students of policy implementation in the American state. While the concept of American state fragmentation has been given a wide range of meanings by scholars, one key idea Wilson s many brawling participants is simply this: implementation authority is splintered and distributed over a larger number of discrete actors and divided over more separate institutions, thereby increasing challenges of coordination and multiplying opportunities for conflict. While this idea of fragmentation has been in circulation for more than a quarter century, in recent years there has been a surge in scholarly interest in its causes and consequences (Freeman and Rossi 2012; Biber 2012; Stephenson 2011; Bradley 2011; Marisam 2011; Gersen 2006; Buzbee 2005; Ting 2003; Kagan 2001). In the domain of statutory policy, this work recognizes that fragmented legal frameworks for policy implementation are a function of legislative design, with legislators assigning roles and powers to actors and institutions that carry a law into effect. One broad question this work has taken up, at a theoretical level, is why legislators would choose fragmented implementation designs. However, despite the level of scholarly interest in the causes of fragmentation, systematic empirical work on the subject is extremely thin. We are aware of no published work that has sought to measure the number of discrete actors, separate institutions, or episodes of overlapping jurisdiction incorporated into policy implementation designs. We argue that important empirical work on legislative designs that constrain administrative or presidential authority in implementation (Epstein and OHalloran 1999; Lewis 2003; Huber and Shipan 2

3 2002) measures different concepts than fragmentation, and that no inferences about fragmentation (as we define it) can be drawn from that work. Absent such a measure, we lack systematic information about basic questions such as whether and how this dimension of fragmentation in the American state has changed over time, and what factors explain its presence, persistence, and patterns of development. Those questions motivate this study. Focusing on regulatory legislation, we develop a novel measure of fragmentation in policy implementation, collect data on it from 1947 to 2008, and test hypotheses linking separation of powers structures to legislative design of fragmented implementation power. We are centrally interested in the effect of divided party government on fragmentation, and the theory that Congress uses fragmentation to inhibit presidential subversion of congressional preferences in implementation. We demonstrate that divided party government is powerfully associated with fragmentation, and that this association contributed to the long-run growth of fragmentation in the post-war U.S. We further find that legislative coalitions are more likely to fragment implementation power amid greater uncertainty about remaining in the majority. Fragmenting Implementation We first advance a conceptually clear definition of fragmentation. We discern in the literature a number of distinct, while clearly interrelated, dimensions of the concept of fragmentation that we wish to integrate. First, more fragmented policy implementation designs rely upon a larger number of distinct actors and entities such as boards, commissions, secretaries, separate administrative officers, judges, and litigants to carry a law into effect (Wilson 1989; Kagan 1999, 2001; Jordana and Sancho 2004). On this view, the sheer volume of actors and entities playing a role in implementation contributes to its fragmented character. Second, power can be fragmented by dividing it over multiple distinctive sources of institutional authority, each of which has a significant measure of autonomy and independence. 3

4 This vein of work, as applied to the United States, has largely focused on the distribution of implementation power across separate administrative agencies (Freeman and Rossi 2012; Biber 2012; Bradley 2011; Gersen 2006; Buzbee 2005). This concept is distinct from the number of separate actors and entities since the sheer number does not register whether such actors and entities are located within a single agency or across multiple agencies. The literature suggests, and it is sensible to expect, that dividing power across multiple agencies raises distinct challenges of coordination because separate agencies can have different even conflicting policy goals, priorities, and ideologies (Clinton and Lewis 2008). Third, a distinctive form of fragmentation arises when legislators not only empower multiple actors and/or agencies to implement policy, but also give them authority and responsibility to perform the same functions with respect to the same statutory provisions, such as empowering two different administrators to make rules on the same subject (Freeman and Rossi 2012; Bradley 2011; Marisam 2011; Gersen 2006; Ting 2003). Scholars have characterized this form of fragmentation as redundant, duplicative, or overlapping by design, potentially instigating turf wars. This scenario is distinct from assigning multiple actors and/or agencies separate functions to be performed in concert, each with exclusive jurisdiction over its delegated function. Drawing these threads together, we characterize a design as highly fragmented if it relies upon many actors, numerous agencies, and contains frequent episodes of overlapping jurisdiction. Below, we propose measures of each of these three forms of fragmentation, and a composite measure that taps the underlying concept that unites them. Why Fragmentation Matters Scholars working from multiple perspectives have long regarded fragmentation s consequences as vitally important, though they have reached radically different conclusions about its effects. One line of analysis emphasizes that fragmentation imposes great costs on American society by producing a structurally ramshackle and operationally dysfunctional state. 4

5 It makes government inefficient, ineffective, unresponsive, and incoherent, and produces legal uncertainty, indeterminacy, and contradiction (Moe 1989, 1990; Schuck 1992; Kagan 1999; 2001; Lewis 2003; Marisam 2011). Because fragmentation weakens control over policymaking by elected officials, it erodes the democratic accountability of policymakers (Moe 1989, 1990; Lewis 2003, 3; O Connell and Farber 2014). It renders law and policymaking excessively complex and opaque, biasing outcomes in favor of elites endowed with legal and political resources needed to navigate it, thus delegitimizing the state in the eyes of the wider public (Schuck 1992). However, a different theoretical account of fragmentation has grown in recent years, emphasizing its potential virtues. On this view, delegation to more actors and institutions can produce competition and innovation that advances state efficiency and effectiveness; cultivate the creation and use of valuable information and expertise; productively leverage distinctive forms of institutional capacity; and foster the representation of a wider range of groups and interests in the policymaking process (Landau 1969; Busch, Kirp, and Schoenholz 1999; Ting 2003; Gersen 2006; Stephenson 2011; Biber 2012). It can also insulate agencies from capture because of the added difficulty associated with capturing many implementers (Bradley 2011: 778). It can aid congressional monitoring of implementers by creating a system of inter-agency fire alarms, with some implementers incentivized to surveil and report on others to Congress, thus enhancing democratic control and accountability (Freeman and Rossi 2012: 1138). Debates over the consequences of fragmentation are empirically and normatively complex. They are also unsettled. These debates are all the more salient amid increasingly fragmented regulatory designs in the contemporary era, which we document below, and which may explain the proliferation of work on fragmentation in recent years. A paper seeking to shed new light on the causes of fragmentation cannot adjudicate among contending perspectives on its consequences. However, the rich debate over the consequences of fragmentation for 5

6 American governance and democracy highlights the importance of understanding where it comes from and how it has changed over time. Fragmentation, Divided Government, and Policy Control Divided government creates incentives for Congress to fragment implementation power, and the postwar growth in the conditions of divided government has importantly contributed to the material increase of fragmentation in federal regulatory policymaking. Divided party government contributes to fragmentation through the mechanism of legislative-executive conflict over control of bureaucracy. Legislators and the interest groups that influence them are aware that presidents possess considerable capacity to influence agency behavior, and they design laws to guard against that influence (Moe 1989, 1990; Epstein and O Halloran 1999; Huber and Shipan 2002; Lewis 2003). Divided government greatly increases an enacting coalition s attention to designing frameworks of policy implementation meant to achieve its own policy goals while insulating from executive subversion. Huber and Shipan (2002) find that divided government leads to more detailed laws, with detail measured by a law s word count. Epstein and O Halloran (1999) find that divided government leads Congress to delegate less discretion to the bureaucracy, with lower degrees of discretion measured by higher levels of formal structural constraints on administrative action (e.g., appropriations limits, reporting requirements, or sunset provisions) relative to the magnitude of the tasks delegated. Lewis (2003) finds that when creating new agencies under divided government, Congress is more likely to structurally insulate the agency from presidential influence through mechanisms such as imposing qualifications on who the president can appoint, fixing the duration of their service, and placing agencies at a greater remove from presidential control. This body of work amply demonstrates that legislative coalitions facing an ideologically opposed executive branch strategically design legislation with the goal of guarding against policy shifting away from the enacting legislative coalition s preferences. We stress that none of the foregoing studies examined whether divided government is 6

7 associated with fragmentation as we have defined it: division of implementation authority over a larger number of distinct actors, over a larger number of different agencies, and giving multiple actors the authority to perform the same function with respect to the same statutory provisions. Nothing can be inferred about fragmentation, so defined, from the dependent variables in this earlier work. As a logical matter, it is possible that laws can be very detailed as measured by Huber and Shipan, delegate little discretion as measured by Epstein and O Halloran, and create highly insulating agency structures as measured by Lewis, while lodging all power to implement a law in a single institutional actor (such as a Secretary or a Board), sitting within a single agency, with no episodes of overlapping jurisdiction. Conversely, a vague statute, with few express constraints on bureaucratic discretion, and few attributes of structural agency insulation, could spread implementation over numerous institutional actors and numerous agencies, and contain many episodes of overlapping jurisdiction. Below, we compare our measure of fragmentation to Lewis s measure of insulation and find that the two are unrelated as an empirical matter. 1 Fragmentation and Constraint. Fragmentation of an implementation framework can serve the enacting coalition s goal of constraining implementers from subverting its preferences, contributing to the stickiness of the status quo. This feature of fragmentation will be more attractive to coalitions under divided government. Increasing the number of actors and agencies that must be coordinated to accomplish decisive action can, on balance, 1 We note that while Epstein and O Halloran (1998) and Huber and Shipan (2002) do not discuss fragmentation, Lewis (2003) does. However, Lewis is referring to a phenomenon quite different and very much broader than our concept of fragmentation. While Lewis s concept includes placing multiple agencies in competition and creating overlapping jurisdictions (8, 27-28), it further includes all things that significantly constrain the president s ability to manage the bureaucracy (4), or that limit political control of agencies (10). Thus, Lewis considers allowing a single independent regulatory commission to dominate policymaking on an issue despite its highly unitary and concentrated power to be an example of fragmentation because it limits presidential influence relative to an agency headed by a cabinet secretary (22). Likewise, because more specific and detailed statutes are associated with lesser presidential discretion with Congress determining more policy in the statute Lewis characterizes them as fragmenting (10). His concept of fragmentation thus embraces all that limits presidential discretion. We use the same word to denote a much narrower concept: the splintering of power over a larger number of distinct actors and different agencies, and more episodes of overlapping jurisdiction among them. 7

8 make significant departures from the policy status quo more difficult. It creates coordination challenges (Freeman and Rossi 2012: 1138) and a system of checks and balances (Jordana and Sancho 2004: 302) that will limit presidential influence on implementation of the policy in question (see also Moe 1989; Williamson 1993; Lewis 2003). For this reason, presidents have consistently favored more unified, centralized, hierarchical administrative designs, which give them more power over implementation (Moe 1989; O Connell 2006: 1704; Lewis 2003). Of course, a fragmented implementation design will also constrain the enacting coalition itself if it seeks to manage bureaucratic drift arising from bureaucrats own preferences, interests, and objectives, which can be quite different from their legislative or executive principals (McNollgast 1987; Moe 1989). However, despite this reduction in control by the enacting coalition, there are several reasons why the calculus will still often favor fragmentation under divided government. First, ideologically distant presidents with hierarchical control pose a greater threat to the enacting coalition s preferences than bureaucratic drift arising from other sources (Moe 1989; Moe and Howell 1999; Lewis 2003). Presidents in American government are held distinctively responsible by their support coalition for the output of federal policymaking institutions. They also possess unique institutional capacity for unilateral action that Congress has difficulty countering. Thus, when presidents have distant policy preferences and are armed with robust, hierarchical instruments of control, they will have both strong political incentives and potent institutional capacity to depart from the enacting coalition s preferences. Under these conditions, hostile presidents are rightly regarded by enacting coalitions as the most fearsome threat of subversion on the American political landscape (Moe 1989: 281). Second, when Congress uses fragmentation to guard against subversion by distant presidents, it is not left without means to manage bureaucratic drift arising from bureaucrats own preferences. It can use administrative procedures and constraints to control bureaucracy, such as spending limits, time limits, public hearing requirements, legislative vetoes, 8

9 and appeals procedures (Epstein and O Halloran 1999: ; Wilson 1989: 243; McCubbins and Schwartz 1984). Decisions about agency funding through appropriations bills provide another particularly potent mechanism of controlling bureaucratic drift arising from bureaucrats own preferences (Davidson and Oleszek 2006: ; Eskridge et al. 2014: ; Headrick et al. 2002). The use of public hearings and investigations can also subject recalcitrant bureaucrats to public humiliation that devastates their careers (McNollgast 1987: 249; Davidson and Oleszek 2006: 355). Thus, while it is surely true that fragmentation, ceteris paribus, can limit the enacting coalition s ability to use hierarchical controls to manage bureaucratic drift, it can still be a sensible strategy under divided government. This is because: (1) ideologically distant presidents with strong instruments of hierarchical control pose a distinctively worrisome threat, and (2) the enacting coalition has other means at its disposal to address the problem, albeit imperfectly. Fragmentation and Inducement. Moreover, an old but recently burgeoning line of work emphasizes that, in addition to simply hobbling presidential control of the bureaucracy, fragmentation may serve a number of additional legislative goals. As discussed above, this work argues that placing multiple actors in the same policy space can engender productive competition and innovation; produce better information; and harness distinctive forms of institutional capacity (Landau 1969; Busch, Kirp, and Schoenholz 1999; Ting 2003; Gersen 2006; Stephenson 2011; Biber 2012). It can mitigate the risk of agency capture by interest groups hostile to the enacting coalition s preferences because it is more difficult to capture many implementers (Bradley 2011: 778). It can contribute to congressional capacity to monitor implementers by creating interagency fire alarms, with some implementers monitoring and reporting upon others to Congress (Freeman and Rossi 2012: 1138). The use of such institutional design strategies to induce bureaucratic behavior consistent with the enacting coalitions preferences will be more attractive when it sees the executive as an enemy, prone to lead the bureaucracy in the wrong direction. Moreover, this point is related to the cost 9

10 that Congress faces when it fragments, thereby weakening its own capacity for hierarchical control. The notion that fragmentation may induce desired bureaucratic behavior suggests that it may thereby reduce the need for congressional supervision through hierarchical control. In sum, fragmentation imposes added burdens of inter- and intra-agency coordination on a president from the opposing party seeking to control policy implementation. It can also affirmatively induce implementation behavior desired by Congress when a distant president threatens to subvert. Thus, we predict that divided government will be associated with higher levels of fragmentation. Divided Government Hypothesis: Under conditions of divided government, Congress will be more likely to fragment implementation power. Other Institutional Sources of Fragmented Design of Implementation Power Existing literature suggests a number of other political-institutional theories to explain fragmentation. An adequate empirical model of the relationship between divided government and fragmentation must incorporate these theories, and given that they have not previously been tested, they are of considerable interest in themselves. Legislative Bargaining and Fragmentation. It has been argued that fragmentation in implementation arises from political compromise, with fragmentation being used to buy the support of political opposition (Moe 1989; King 1997; Ting 2003; Lewis 2003; see also Maltzman and Shipan 2008). To achieve passage of legislation amid partisan conflict, winning legislative coalitions must usually compromise with the losing group when making decisions about bureaucratic structure, and the losing group is dedicated to curtailing agency powers and gaining influence over agency decisions in whatever ways it can. Thus, the losing group will pressure for fragmented authority with the goal of constraining administrators and hampering vigorous implementation by the opposing party (Moe 1989: 326). This institutional bargaining logic suggests two testable hypotheses. First, the greater the 10

11 divisiveness and opposition in the legislative politics surrounding a bill, the more likely that fragmentation will be used as a strategy to buy the support of political opposition. Divisiveness Hypothesis: As the legislative divisiveness surrounding a legislative proposal increases, it will be more likely to fragment implementation power. Second, legislative majorities governing by thin margins of control will be more likely to fragment implementation power than will majorities who govern by wide margins of control. The thinner a majority s margin of control, the greater will be its need to compromise in implementation design in order to achieve passage. Margin of Control Hypothesis: As the majority s margin of control becomes thinner, it will be more likely to fragment implementation power. Electoral Uncertainty, Coalition Drift, and the Stickiness of the Status Quo. Moe (1989) argues that current legislative coalitions are mindful of the prospects of electoral defeat and the possibility that future legislative coalitions will seek to guide implementation in an undesirable direction. This is the problem of coalition drift. Congress thus enacts formal rules and structures calculated to limit bureaucratic discretion with the goal of insulating policy from future legislative coalitions. This strategy of insulation can be effective in the American separation of powers system because the many impediments that the system presents to enacting laws (particularly its many veto points) often apply with even greater force to repealing an existing law, around which vested interests may already have formed (Moe 1989). This institutional logic for constraining bureaucracy operates as an incentive to fragment implementation power. As Oliver Williamson characterized this cause of fragmented bureaucratic design, Incumbent politicians who create and design bureaus are aware that the opposition can be expected to win and take control in the future, and thus a farsighted majority party will...design some degree of (apparent) inefficiency into the agency at the outset...to frustrate the efforts of successors...to reshape the purpose served by the agency (Williamson 1993: 107). According to this logic, greater concern about loss 11

12 of power by the majority party will make an enacting coalition more likely to fragment. Electoral Uncertainty Hypothesis: As the risk of losing power through elections increases for the majority party, it will be more likely to fragment implementation power. The Data and Model The Body of Laws In order to build a dataset of laws to investigate the fragmentation of implementation power, we start with the 366 federal statutes, passed between 1947 and 2008, that were identified by David Mayhew (1991) as highly significant. 2 We study implementation power in the context of domestic regulation. Coders read all of the Mayhew laws in order to identify those containing domestic regulatory commands. Regulatory, as used here, refers to any governmental effort to control behavior by other entities, including...business firms, subordinate levels of government, or individuals (Foreman 2001: 12982). A regulatory command is a mandatory proscription of actions that the legislation seeks to prevent (e.g., an employer must not discriminate based upon race), or a mandatory requirement that the regulated population engage in required conduct (e.g., a lender must disclose required information to a borrower). Using this broad conceptualization of regulatory commands, we included all laws that contained any civil regulatory commands directed at behavior within the United States. We emphasize the breath of our concept of regulatory policy. In addition to traditional regulatory policy, our data includes commands that are a part of welfare state policy, as with, for example, commands meant to regulate Medicaid fraud and the use of food stamps. It also includes commands aimed at controlling the expenditure of federal funds for wide-ranging purposes, including transfer programs, infrastructure building, and economic development. Our choice to limit the study to regulatory policy, so defined, was guided by the judgment that it would allow us to achieve a consistent measure of fragmentation. Theoretically, any 2 Mayhew has provided updates to his identification of significant legislation over time. 12

13 mandatory command implicates a comparable range of implementation options, and thus is susceptible to the same menu of types of delegations. Further, and critically to our design, this limitation allows a consistent measure of the magnitude of the intervention, which we regard as a critical control. We address this further below when we describe the data and how it was coded. In the conclusion, we consider the prospects for extending the paper s insights into other broad policy types. Of Mayhew s 366 significant statutes, 218 contained domestic regulatory commands. The average page length of the 218 laws was 110 pages, and thus the federal statutory law that is the basis of the analysis that follows spanned approximately 23,980 pages of the Statutes at Large. Each law was read in full in order to code the variables described below, which demanded identification of detailed substantive information about the policy content of each statute, as well as detailed information about the provisions governing implementation of the statute s regulatory commands. Measuring Fragmentation: The Dependent Variable Actors Count. We counted each discrete named actor/entity in each law that was empowered to execute one or more core regulatory functions to implement a law s regulatory commands, which we define as (1) holding adjudications, (2) making administrative rules, (3) imposing some type of administrative sanction or order, or (4) prosecuting enforcement actions. Common actors delegated administrative implementation authority are Secretaries, Administrators, Boards, Offices, Commissions, Commissioners, or Directors working within a federal administrative body. In addition to administrative actors, we counted judges as actors when the law relied in part on direct enforcement through lawsuits, and we counted private prosecutors as actors if the law contained a private right of action. Federal courts and private litigants are fundamentally important parts of the implementation of federal statutes (Farhang 2010), and they have been incorporated into important conceptualizations of fragmentation in American policy implementation (Wilson 1989: ; Moe 1989: 276; 13

14 Kagan 2001). Agencies Count. We counted the number of different federal agencies delegated some authority to carry out a core regulatory function to implement regulatory commands in the law. Included in the agencies count are executive branch agencies, independent administrative agencies, agencies of the legislative branch, and federally-created for-profit corporations that served the role of implementing the act s regulatory commands. Overlapping Functions Count. We counted the number of times that multiple actors were simultaneously given the authority to perform the same regulatory implementation function in order to implement the same regulatory commands of a law, such as where two administrators are empowered to make rules on the meaning of the same commands, or to prosecute lawsuits alleging violation of the same commands. This applies narrowly to actors implementing the same regulatory commands of a law, and not simply to actors given the same type of functional authority to be exercised over different regulatory commands. We provide examples of overlapping functions from our data in the Appendix. Table 1 contains examples of the application of our coding protocol to portions of laws in our data. The implementation provisions appear on the left side of the table, and the corresponding counts for actors, agencies, and overlapping functions appear on the right side of the table. In the Appendix, we provide further discussion of how we coded these examples. Fragmentation. Each of the above measures taps a different dimension of fragmentation. A variable that combines information from each is superior to any one measure standing alone. The dependent variable of fragmentation is, for each law, the mean of the standardized measures for actors, agencies, and overlapping functions. To test the validity of this scale, we first computed Cronbach s alpha, which measures the internal consistency of a scale, or the extent to which multiple items in a scale measure the same underlying concept or construct. The Cronbach s alpha for these variables was.9, which is considered to be 14

15 excellent validity (DeVillis 2012; Kline 2000). We also used principal components factor analysis with orthogonal rotation to compute a factor score for each law based upon the three measures. The variables loaded heavily on one factor, rendering an eigenvalue of 2.5 for the first factor. The factor loadings on the first factor for the three items range between.85 and.96, showing that all three items are strongly associated with the underlying factor, and at a roughly comparable level. The second largest eigenvalue was only.4. The very strong single-factor outcome was clearly confirmed by examining a scree plot. The fact that all three items load heavily on one factor again shows, consistent with the Cronbach s alpha, that the three items comprising our fragmentation scale are measuring the same underlying construct in an internally consistent and reliable fashion (DeVellis 2012). The factor score generated was correlated with the unweighted mean of the standardized variables at.999, and performed virtually identically in the models presented below, with no meaningful differences in statistical or substantive significance. Consequently, in our main analysis we display and discuss the simpler and more transparent approach of the unweighted mean of the standardized variables. We display parallel models using the factor score as our dependent variable in Table A1 of the Appendix. Figure 1 displays lowess plots of the composite measure of fragmentation from 1947 to 2008, as well as for each of the three underlying fragmentation measures. The composite fragmentation measure reveals that fragmentation in significant federal statutes was relatively flat during the first two decades of the series; grew significantly beginning in the 1960s; roughly plateaued in the mid-1980s; and experienced further growth over the last decade of the series. All three underlying fragmentation measures followed a similar developmental path. This clear commonality over time is consistent with the excellent Cochran s alpha score and the heavy loading on a single factor, further reinforcing the internal consistency of the composite measure. Actors, agencies, and overlapping functions all cohere in 15

16 tapping a single underlying dimension of fragmentation in legislative design. In order to compare our fragmentation variable to important existing work on legislative strategies of agency constraint, we examine Lewis s (2003) agency insulation data. 3 We constructed an additive index from his six measures of agency design attributes calculated to insulate from presidential control. The correlation of the average annual values of our fragmentation index and the index of Lewis s insulation measures is.012; the index of Lewis s insulation measures does not have a positive (or negative) time trend, and the movement of the two measures over time appears unrelated; and there is no statistically significant relationship between them. Details of the comparison are provided in the Appendix. The comparison makes clear that our fragmentation measure taps something different than agency designs using express structural limits on presidential influence. Political-Institutional Variables Divided government is coded 0 when the president s party controls both chambers of Congress, and coded 1 otherwise. Figure 2 shows differences in the mean values of the composite fragmentation scale, and the three underlying measures, under divided versus unified government. With respect to all four, there is a clearly higher mean value under divided government. Relative to its value under unified government, the fragmentation scale increases by 38 percent under divided government. To test the theory that more divided Congresses will be more likely to engage in bargaining that leads to fragmentation, we include both law-level and Congress-level measures of congressional divisions. To test the divisiveness hypothesis, we include a law-level partisan divisiveness variable. Following Harbridge s (2015) approach to measuring partisanship in rollcall voting, we subtracted the percentage of Republicans voting yea (0-100) from the percentage of Democrats voting yea (0-100). This yields a variable ranging from -100 to 100, with -100 indicating a vote on which no Democrats voted yea and all Republicans voted yea, and 100 indicating a vote on which 3 Epstein and OHalloran s data are not publicly available, and Huber and Shipan did not examine federal legislation. 16

17 all Democrats voted yea and no Republicans voted yea. We then took the absolute value. Thus, the variable is 0 where the identical percentage of Democrats and Republicans voted yea, indicating perfect bipartisanship. It is 100 when there is perfect party line voting, or the greatest degree of partisanship in voting. From 0 to 100 there is a continuous movement from complete bipartisanship to complete partisanship. 4 At the Congress level, to test the margin of control hypothesis we include margin of control, which is defined in each chamber as the number of seats held by the majority party minus the number of seats held by the minority party, divided by the total number of seats, averaged across the two chambers. To facilitate an interaction discussed below, we invert the variable s direction by subtracting it from 1. Thus, higher values reflect more narrowly-divided Congresses, which theory predicts will be associated with higher levels of fragmentation. To test the hypothesis that Congress will be more likely to fragment implementation power in the face of electoral uncertainty, we measure close races in the last round of elections, defined as the percent of total seats in the last national election that were won by five percentage points or fewer. We believe that having had more tight races in the prior election cycle, as compared to fewer, is associated with higher levels of perceived electoral vulnerability. Because uncertainty is difficult to measure, as a robustness check we also utilize an alternative variable measuring actual seat share losses in the next election, defined as 4 An alternative view may be that more consensual votes, rather than more conflictual votes, will be associated with greater fragmentation because consensual votes occur as a result of accommodation of the minority through fragmentation. However, we believe that the divisiveness hypothesis predicts a positive association between divisive votes and fragmentation for the following reason. Our data are limited to laws that passed. We anticipate that where accommodation of minority preferences was necessary to passage, the majority would accommodate sufficiently to achieve passage, but not more. In this scenario, within the set of laws that passed, the most divisive votes (those that achieve just enough consensus to pass) would be associated with fragmentation, and the least divisive votes (unanimous and near unanimous ones) would occur in scenarios in which fragmentation to achieve passage was either not necessary, or less necessary. We discuss the issue further in the interpretation section. In alternative specifications, we also substituted a measure of divisiveness in the vote that did not account for party: the percentage of all legislators voting against each law within each chamber, averaged across the two (see Maltzman and Shipan 2008: 260). This variable provided insignificant, as did partisan divisiveness. Like Maltzman and Shipan, we treat voice votes as having no votes against the law (2008: 260, n. 18). 17

18 the percentage of seats in each chamber, relative to total seats, lost by the majority party, averaged across the two chambers. Seat share losses for the majority party takes on positive values when the majority loses seats, and negative values when it gains seats. Further, we model the possibility that the effects of electoral uncertainty on fragmentation will be conditional upon margin of control. Uncertainty among the majority coalition about how it will fare in the next election may be especially worrisome to a party holding on to power by a thin majority, when the loss of only a small number of seats would move them to minority status, and this will heighten their incentive to insure against future loss of power by inducing status quo bias. On the other hand, when holding power by a wide margin of control, signs of electoral uncertainty may be, by comparison, less likely to induce efforts to insure against loss of power. Thus, the interaction of the electoral uncertainty variables defined above and margin of control captures uncertainty about remaining in the majority. The electoral uncertainty variables and margin of control were centered on their mean to facilitate interpretation of the interaction. We also account for the possibility that partisan preferences are associated with fragmentation. It is commonly believed that liberalism in American politics is, other things equal, associated with greater support for centralized state control, while conservatism is associated with greater skepticism of a centralized, hierarchically organized national government. To control for the possibility that Republican Congresses will be more likely to fragment implementation power than Democratic Congresses, we include partisan seat share, which is positive when Democrats control Congress and reflects their margin of total seats over Republicans in each chamber, divided by total seats in each chamber, averaged across the chambers. It is negative when Republicans control Congress and reflects their margin over Democrats. Law Characteristics Policy Domain. The nature of policy areas regulated in a statute may be associated 18

19 with our fragmentation measures given, for example, their degree of complexity. We assign each law a policy code and included a set of fourteen policy area dummy variables. Scope. We control for the magnitude or scope of the regulatory interventions undertaken in a law since more extensive regulatory interventions may require more actors and agencies, and present more opportunities for overlapping functions. To measure the extensiveness of a statute s intervention, scholars have developed a protocol for counting a statute s major provisions (Franchino 2007: 109; see also Epstein and O Halloran 1999: 275). We adapt this approach to the task of counting regulatory commands in each law. The core idea is to count each separate regulatory command, resulting in a sum of the number of regulatory commands, or the requirements and prohibitions imposed upon regulated entities. The regulatory commands measure does not count material amplifying upon the meaning of the command, such as definitions, exceptions, rules of application, or rules of construction. Such information concerns a law s level of detail, not the scope of its intervention. The regulatory commands measure also does not count portions of the statute allocating and defining authority delegated to implementing actors. Such information importantly affects our dependent variable, which is based upon counts of delegated authority. Rather, based upon content analysis of each law, regulatory commands isolates and measures the number of discrete commands of substantive regulatory policy undertaken in the law. Complexity. Growth in policy complexity has been identified by some as an explanation for the growth in fragmentation of administrative power (Schuck 1992; Barnes 2004: 35). That is, the intrinsic complexity of some policy issues leads to more fragmented implementation designs. One approach in the literature to measuring statutory policy complexity is to use the length of statutes in pages or words (Franchino 2000; Maltzman and Shipan 2008). Maltzman and Shipan theorize that policy complexity is associated with laws touching upon a larger number of issues, and they use page length as a proxy for this. To a considerable extent, this idea of complexity overlaps our regulatory commands variable, which is a 19

20 more direct and targeted measure of the number of issues regulated. However, there may be aspects of the complexity of a regulatory law s intervention other than number of issues regulated, and thus we incorporate page length as well. An additional approach to measuring policy complexity focuses on the nature of the issues regulated rather than the scope or number of issues regulated. Franchino (2007: 145) measures complexity of legislative subject matter based upon whether legislation contains a rulemaking delegation, arguing that rulemaking delegations are associated with the legislature wishing to avoid the investment of time necessary to make policy in complex areas, instead leveraging administrative capacity. With respect to each regulatory command, we recorded whether an administrator was delegated substantive rulemaking authority covering that part of the law. Generally, such rulemaking authority exists only if it is specifically delegated (Pierce 2010: ). We then constructed rulemaking ratio, which is the number of regulatory commands in each law governed by delegated substantive rulemaking authority, divided by the total number of regulatory commands in each law. Higher values are associated with greater reliance on agency rulemaking. Time Trends Time trends, and the potentially dynamic nature of fragmentation in the postwar U.S., must be addressed for several reasons. Figure 1 reflects that fragmentation grew strongly in the 1960s, roughly plateaued in the mid-1980s, and grew again at the end of the series. Over the same period, divided government became more frequent. Thus, we must account in the model for the broad growth trend in fragmentation in order to confidently isolate the effects of divided government. Further, as a theoretical matter, it is possible that choices by Congress concerning fragmentation at a given point in time will be affected by fragmentation levels in earlier periods. Past levels of fragmentation may encourage choices to fragment in subsequent periods by increasing the number of already existing actors who may receive delegations. Fragmentation may also beget fragmentation to the extent that it reflects changing norms of 20

21 regulating in Congress. When coalitions draft laws, they may be influenced by recent models of implementation designs, whether unitary or fragmented. We incorporate a measure of general growth in the administrative state over time. The sheer growth in executive branch size may be associated with growth in the number of potential recipients of delegations, and thus opportunities to fragment implementation. Executive branch size reflects total executive branch civilian employment. 5 In one set of model specifications, we include year, which is a simple linear time trend to model the growth of fragmentation over time. In alternative specifications, we model the nonlinear growth and development of fragmentation using standard cubic spline functions with knots to fit the shape of changes in fragmentation over time. 6 When using non-linear splines to model a time trend, it is appropriate to place knots in the series at points of inflection (Eubank 1999: ), and we rely upon the representation of the data in Figure 1 to identify them. The points of inflection are at 1960, 1985, and The temporal function created by the procedure is defined as a continuous smooth function that is linear before the first knot, is a piecewise cubic polynomial between successive knots, and is again linear after the last knot. The procedure always creates one fewer variables than knots (Calcagno, Hu, and Kanigel 2013). This estimation strategy allows us to explicitly model the pattern of growth of our composite measure of fragmentation over time. Findings Descriptive statistics are presented in Table 2. Our dependent variable measuring fragmentation is continuous and thus we analyze it using ordinary least squares regression. In each of the models below, we cluster standard errors by Congress to allow for correlation within Congresses in our data. Model A in Table 3 presents a sparse model with only our independent variables testing the institutional hypotheses, along with the linear time trend. 5 The data come from Historical Federal Workforce Tables, Office of Personnel Management. 6 We use Stata s rc spline command to create the spline variables. 21

22 Model B adds other control variables. Model C substitutes the cubic spline for the linear time trend. The key results are consistent across all three models. Divided government is significant with the expected sign in all three models, and the magnitude of the effect is somewhat reduced by inclusion of the controls. To put the magnitude of the effect in perspective, consider that the mean value of our fragmentation scale under unified government is The coefficient on divided government in the full model indicates that conditions of divided government are associated with an increase in the fragmentation scale of.38, or 36 percent over its mean under unified government. In Model C, the increase is 33 percent over the mean under unified government. Partisan divisiveness is insignificant, suggesting that the extent of partisan divisiveness over a law is not associated with higher levels of fragmentation. We also considered, as an alternative operationalization of divisiveness, the share of total votes that were nay, averaged across the chambers. This proved insignificant as well. It is possible that this lack of a divisiveness effect arises from the fact that concessions making a law more fragmented also have the intended effect of achieving a more consensual vote, whereas divisive votes occur where the majority coalition made the fewest concessions to the minority. 7 Even if this dynamic were at play in rollcall behavior, the margin of control hypothesis would still predict that Congresses with majority parties governing with thinner margins of control would pass laws with more fragmented implementation. While the evidence would not exist at the individual rollcall level, the fragmentation-enlarging concessions made to achieve more consensual votes would be increasingly necessary in more closely divided Congresses. However, the main effect of margin of control is also insignificant in all three models, indicating that the margin by which the majority party controls Congress is not associated with higher levels of fragmentation. The main effect of close races is insignificant as well, indicating that the percent of close races in the last round of elections was not associated 7 See note 4 for a brief discussion of why we nevertheless think that it is theoretically sensible to predict a positive relationship between divisive votes and fragmentation. 22

23 with fragmentation. Because margin of control and close races are centered on 0 and interacted, the coefficient and standard error on each variable reflects the results only when the other variable is at its mean. 8 However, the interaction of margin of control and close races is significant in all three models, with the magnitude of the effect somewhat reduced by inclusion of the controls. This variable captures electoral uncertainty not just in terms of the number of competitive races in the last round of elections, but also conditional upon the majority party s margin of control, with low values associated with wide margins of control and few competitive races, and high values capturing thin margins of control and many competitive races. In the full model with a linear time trend (Model B), an increase of one standard deviation in the interaction is associated with a.16 unit increase in fragmentation. This is an increase of 12 percent over the mean value of the fragmentation index in the data (which is 1.3). Figure 4 presents a plot of this interaction term s effect on fragmentation with 95 percent confidence bands. The figure reveals a clear upward trend, consistent with the theory that under higher levels of uncertainty about remaining in power, Congress will be more likely to rely upon fragmented design for policy implementation. We note that above the value of about.0025 on the interaction, denoted by the vertical line, the lower confidence band reflects that the upward slope cannot be distinguished from a zero slope at the 95 percent confidence level. However, 90 percent of our data falls below that level. The portion of the figure above.0025 is very sparsely populated with data, resulting in uncertain estimates. In the region below.0025, where 90 percent of our data lie, the effect is clearly positive and statistically significant. In the spline model, the effect of a one standard deviation increase in the interaction is also associated with an increase of 12 percent in fragmentation over the mean value of the fragmentation index. In order to test the robustness of the electoral uncertainty finding, we substitute seat 8 Margin of control and close races remain insignificant even if their interaction is dropped from the model. 23

24 share losses for close races and rerun all three models. The results are presented in Table 4. The main effect of seat share losses is significant in every model with the expected sign. An increase of one standard deviation in seat share losses is associated with a.09 increase in fragmentation, or an increase of seven percent over the mean value of the fragmentation index, in both the full model with the linear time trend (Model B), and the spline specification (Model C). The interaction of seat share losses and margin of control is insignificant in all three models. Despite this inconsistency in the conditionality of the alternative measures of electoral uncertainty upon margin of control, in both cases we observed quite differently constructed measures of electoral uncertainty to be significantly and positively associated with fragmentation. With the alternative uncertainty measure in the linear model (Model B), divided government is associated with a 26 percent increase in fragmentation relative to its mean value under unified government. The increase is 24 percent in the spline model (Model C). Partisan seat share is insignificant. Higher levels of Republican control of Congress are not associated with more fragmented implementation powers. Executive branch size is also insignificant, such that the sheer size of the executive branch is not associated with the legislative choice to fragment power. Regulatory commands, measuring the scope of the regulatory intervention, is highly significant with a large effect. Across all models with this control in Tables 3 and 4, a one standard deviation increase is associated with an increase of.32 in fragmentation, or 25 percent over its mean. Page length and rulemaking ratio are insignificant. Finally, in the Appendix we present models with individual counts of actors, agencies, and overlapping functions as dependent variables. With respect to all three individual dependent variables, the results on the divided government, divisiveness, margin of control, and electoral uncertainty hypotheses are consistent with those in the composite model. While we view the composite measure as clearly superior in characterizing fragmented leg- 24

25 islative design, the findings are robust to specifications with the individual components of fragmentation. Conclusion A generation of scholars has debated the causes and consequences of the fragmented character of the American state s infrastructure for implementing public policy, with its many participants in competition and conflict. This rich and important literature has lacked an adequate empirical measure of fragmentation to subject theories to systematic evaluation, or even to map its development over time. In this paper, we have advanced a conceptually clear definition of fragmentation; developed a novel empirical measure of this facet of the American state spanning the postwar period; and applied it to a body of important regulatory laws, requiring detailed content analysis of about 24,000 pages of the United States Statutes at Large. Our analysis of this measure reveals robust, consistent evidence that under conditions of divided government, Congress is more likely to fragment regulatory implementation power. The barroom brawl of American policy implementation has grown larger and more intense since the 1960s, and the persistent condition of divided party government has been a critical variable contributing to this escalating fragmentation. American presidents wishing to pursue their own policy missions have inherited structures that are, from their point of view, more and more cumbersome, unwieldy, and hard to manage. They were, in part, intended to be so. Congresses wishing to check subversion of legislative preferences by hostile executives have denied them effective and efficient means to go their own way. Moreover, there is every reason to believe that the fragmented structures thereby created are highly durable. Political scientists have emphasized that legislation can create benefits that mobilize beneficiaries in favor of maintaining a policy in the face of attempts to change it (Pierson 1994; Weir and Skocpol 1985). The multiplicity of newly powerful secretaries, boards, committees, interest groups, lawyers, and litigants can be expected to defend their piece of implementation power 25

26 from consolidation under centralized control. Our results also shed light on the character of the American state in cross-national perspective. In describing American policy implementation, lamentably, as a barroom brawl, Wilson (1989) contrasted it with the relatively more centralized and hierarchal structures typical of other economically developed democratic states, which he thought were more effective. Partly in response to Wilson, Moe (1990: ) suggested that this facet of the American state was a function of the American separation of powers system, as contrasted with parliamentary arrangements. Owing to the much closer identity of legislative and executive preferences and interests in parliamentary regimes, legislators are less prone to design implementation structures with the goal of constraining executive power, and thus they are less likely to fragment. Consistent with Moe s insights, this paper s main result ties American constitutional separation of powers structures to the distinctively fragmented character of American policy implementation in cross-national context. Our data and models also provided the opportunity for the first systematic tests of some longstanding theories about fragmentation. Using two quite different measurement approaches, we find that fragmentation is associated with higher uncertainty about remaining in power, with fragmentation used to insulate against coalition drift. We did not find support for the theory, which appears repeatedly in the literature, that divisiveness over the passage of laws incentivizes Congress to fragment implementation power as a strategy to buy support. Nor did we find support for the related hypothesis that thin margins of control by the majority party, by necessitating bargaining to achieve passage, independently lead to fragmentation. Finally, our novel measure of regulatory fragmentation points the way toward studying fragmentation in the implementation of other types of policy. Another major policy type ripe for inquiry is spending laws, in which the federal government appropriates large amounts of funds to be administered and distributed by governmental and private non- 26

27 governmental entities to implement programs to deliver social benefits, economic development, infrastructure-building, and scientific and technological innovation. We believe that something like our measurement strategy could be ported to this policy context, though it would require clearly conceptualizing and operationalizing the menu of delegations from which Congress draws when it constructs an implementation scheme to distribute funds. This would be a productive avenue for future research. 27

28 References Aberbach, Joel D Keeping a Watchful Eye: The Politics of Congressional Oversight. Washington D.C.: Brookings Institution. Barnes, Jeb Overruled? Legislative Overrides, Pluralism, and Contemporary Court-Congress Relations. Stanford: Stanford University Press. Biber, Eric The More the Merrier: Multiple Agencies and the Future of Administrative Law Scholarship. Harvard Law Review 125: Bradley, Keith The Design of Agency Interactions. Columbia Law Review 111(4): Busch, Christopher, David Kirp, and Daniel Schoenholz Taming Adversarial Legalism. New York University Journal of Legislation and Public Policy 2: Buzbee, William W The Regulatory Fragmentation Continuum. Westway, and the Challenges of Regional Growth. Journal of Law and Politics 21: 323. Calcagno, Juan Carlos, Jian Hu, and Benjamin Kanigel Forecasting and Stress- Testing U.S. Vehicles ABS Deals. Journal of Structured Finance Spring 19: Clinton, Joshua, and David Lewis Expert Opinion, Agency Characteristics, and Agency Preferences. Political Analysis 16: Davidson, Roger, and Walter Oleszek Congress and Its Members, 6th ed. Washington D.C.: CQ Press. DeVellis, Robert F Scale Development: Theory and Applications, 3rd ed. Thousand Oaks: SAGE Publications. Epstein, David and Sharyn O Halloran Delegating Powers: A Transaction Costs Politics Approach to Politics under the Separation of Powers. Cambridge: Cambridge University Press. Eskridge, William, Phillip Frickey, Elizabeth Garrett and James Brudney Cases and Materials on Legislation and Regulation: Statutes and the Creation of Public Policy, 5th ed. St. Paul: West Academic Publishing. Eubank, Randall L Nonparametric Regression and Spline Smoothing, 2nd ed. New York: Marcel Dekker. Farhang, Sean The Litigation State: Public Litigation and Private Lawsuits in the United States. Princeton: Princeton University Press. Foreman, Christopher H Regulatory Agencies, in International Encyclopedia of the Social and Behavioral Sciences, ed. Neil J. Smelser and Paul B. Baltes. Amsterdam: Elsevior Science. Franchino, Fabio Statutory Discretion and Procedural Control of the European Commission s Executive Functions. Journal of Legislative Studies 6:

29 Franchino, Fabio The Powers of the Union: Delegation in the EU. Cambridge: Cambridge University Press. Freeman, Jody and Jim Rossi Agency Coordination in Shared Regulatory Space. Harvard Law Review 125(5): Gersen, Jacob E Overlapping and Underlapping Jurisdiction in Administrative Law. The Supreme Court Review 1: Harbridge, Laurel Is Bipartisanship Dead? Cambridge: Cambridge University Press. Headrick, Barbara, George Serra and James Twobley Enforcement and Oversight: Using Congressional Oversight to Shape OSHA Bureaucratic Behavior. American Politics Research 30: Huber, John D. and Charles R. Shipan Deliberate Discretion? The Institutional Foundations of Bureaucratic Autonomy. Cambridge: Cambridge University Press. Jordana, Jacint and Sancho, David Regulatory Designs, Institutional Constellations and the Study of the Regulatory State, in The Politics of Regulation: Institutions and Regulatory Reforms for the Age of Governance, eds. Jacint Jordana and David Levi-Faur. Cheltenham: Edward Elgar. Kagan, Robert A Adversarial Legalism: The American Way of Law. Cambridge: Harvard University Press. Kagan, Robert A Adversarial Legalism: Tamed or Still Wild? New York University Journal of Legislation and Public Policy 2: King, David Turf Wars: How Congressional Committees Gain Jurisdiction. Chicago: University of Chicago Press. Kline, Paul The Handbook of Psychological Testing. London: Routledge. Landau, Martin Redundancy, Rationality, and the Problem of Duplication and Overlap. Public Administration Review 29: Lewis, David Presidents and the Politics of Agency Design: Political Insulation in the United States Government Bureaucracy, Stanford: Stanford University Press. Maltzman, Forest and Charles R. Shipan Change, Continuity, and the Evolution of the Law. American Journal of Political Science 52(2): Marisam, Jason Duplicative Delegations. Administrative Law Review 63: Mayhew, David Divided We Govern: Party Control, Lawmaking, and Investigations, New Haven: Yale University Press. McCubbins, Matthew, and Thomas Schwartz Congressional Oversight Overlooked: Police Patrols vs. Fire Alarms. American Journal of Political Science 28:

30 McNollgast Administrative Procedures as Instruments of Political Control. Journal of Law, Economics and Organization 3: Moe, Terry M. and William Howell Unilateral Action and Presidential Power: A Theory. Presidential Studies Quarterly 29: Moe, Terry M Political Institutions: The Neglected Side of the Story. Journal of Law, Economics, and Organization 6: Moe, Terry M The Politics of Bureaucratic Structure, in Can the Government Govern?, ed. John E. Chubb and Paul E. Peterson. Washington, D.C.: The Brookings Institution Press. O Connell, Anne Joseph The Architecture of Smart Intelligence: Structuring and Overseeing Agencies in the Post-9/11 World. California Law Review 94: Pierce, Richard J Administrative Law Treatise. New York: Aspen Publishers. Pierson, Paul Dismantling the Welfare State? Reagan, Thatcher and the Politics of Retrenchment. Cambridge: Cambridge University Press. Schuck, Peter H Legal Complexity: Some Causes, Consequences, and Cures. Duke Law Journal 42(1): Stephenson, Matthew C Information Acquisition and Institutional Design. Harvard Law Review 124: Ting, Michael M A Strategic Theory of Bureaucratic Redundancy. American Journal of Political Science 47(2): Weir, Margaret and Theda Skocpol State Structures and the Possibilities for Keynesian Responses to the Great Depression in Sweden, Britain, and the United States. In Bringing the State Back In, eds. Peter Evans, Dietrich Rueschemeyer, and Theda Skocpol. Cambridge: Cambridge University Press, 1985, Williamson, Oliver E Transaction Cost Economics and Organization Theory. Industrial and Corporate Change 2(1): Wilson, James Q Bureaucracy: What Government Agencies Do and Why They Do It. New York: Basic Books. 30

31 Table 1: Counting Fragmentation Implementation Provisions Fragmentation Counts Scenario 1: Voting Rights Act Attorney General Prosecute lawsuits Actor Count Two actors: (1) Attorney General (2) federal judges Agency Count One agency: Department of Justice Overlapping Functions Zero Scenario 2: Telecommunications Act Federal Communications Commissioner Prosecute lawsuits Promulgate rules Private Party Prosecute lawsuits Actor Count Three actors: (1) FCC, (2) private prosecutors, (3) federal judges Agency Count One agency: FCC Overlapping Functions One overlapping function: FCC and private prosecutors both given authority to prosecute lawsuits Scenario 3: Federal Water Pollution Control Act Secretary of Labor Actor Count Hold administrative adjudication Four actors: (1) Secretary of Labor, (2) Promulgate rules EPA Administrator, (3) Attorney General, and (4) federal judges Administrator of EPA Agency Count Prosecute Lawsuits Three agencies: (1) Department of Labor, (2) Promulgate Rules EPA, (3) Department of Justice Attorney General Prosecute lawsuits Overlapping Functions Two overlapping functions: (1) Secretary of Labor and Administrator of the EPA can promulgate rules, and (2) Secretary of Labor and Administrator of the EPA can prosecute lawsuits 31

32 Table 2: Descriptive Statistics Mean SD Min Max Fragmentation Variables Composite Fragmentation Measure Number of Actors Number of Agencies Overlap Factor Score Political Environment Variables Divided Government Margin of Control Partisan Seat Share Electoral Uncertainty Spline Spline Executive Branch Size Law Characteristics Rulemaking Ratio Regulatory Commands Page Length Partisan Divisiveness

33 Table 3: OLS with Composite Fragmentation DV, Std. Errors Clustered by Congress A B C Divided Government 0.499** 0.385** 0.367* (0.172) (0.120) (0.137) Partisan Divisiveness (0.301) (0.315) (0.316) Margin of Control (0.821) (1.127) (1.177) Close Races (1.892) (1.360) (1.371) Margin of Control x Close Races ** * * (15.153) (14.955) (15.140) Year 0.020*** 0.019*** (0.004) (0.004) Spline (0.012) Spline (0.014) Partisan Seat Share (0.857) (0.960) Executive Branch Size (0.000) (0.000) Regulatory Commands 0.010** 0.010** (0.003) (0.004) Page Length (0.001) (0.001) Rulemaking Ratio (0.127) (0.127) Policy FE R N p <.001, p <.01, p <.05 33

34 Table 4: OLS with Seat Share Losses Variable A B C Divided Government 0.412* 0.283** 0.255* (0.167) (0.097) (0.114) Partisan Divisiveness (0.300) (0.291) (0.286) Margin of Control (0.920) (1.016) (1.097) Seat Share Losses 0.881* 0.800* 0.794* (0.383) (0.305) (0.310) Margin of Control x Losses (5.734) (4.631) (4.424) Year 0.016** 0.015** (0.005) (0.004) Spline (0.015) Spline (0.017) Partisan Seat Share (0.837) (1.029) Executive Branch Size (0.000) (0.000) Regulatory Commands 0.010* 0.010* (0.004) (0.004) Page Length (0.001) (0.001) Rulemaking Ratio (0.128) (0.128) Policy FE R N p <.001, p <.01, p <.05 34

35 Figure 1: Fragmentation Over Time Fragmentation Actors Year Year Agencies Overlap Year Year 35

36 Figure 2: Difference-in-Means of Fragmentation by Divided Government Fragmentation Actors Unified Divided Unified Divided Agencies Overlap Unified Divided Unified Divided 36

37 37